Thunderbolt 5 Could Double Bandwidth to 80 Gbps, Intel Leak Suggests

[GuilleA]

I haven't seen any USB4 devices yet. Just the M1 Thunderbolt 3/USB4 host controller, and Intel Thunderbolt 4/USB4 host controllers and Thunderbolt 4 peripheral hubs/docks (which support USB4).

I've only come across one, and it's the OWC Thunderbolt Hub. There's no actual benefits to USB 4 afaik, only minor improvements to Thunderbolt 3.

 

[Val-kyrie]

To be fair, someone has already pointed out that the "single" TB display output supports dual-pane 5k displays, so it's clearly not quite as simple as I'd originally suggested. Still, even if the limitation is partly software that doesn't mean it's not there for a good reason (like, even if you could physically connect 3 displays, could the GPU drive them smoothly, without overheating, even in 4k with non-integer scaling?) There's more to "supports" than "sort-of would work".


I don't think anybody should be surprised if the Mac Pro is the last to transition (if at all) especially the 2019 "if you're not going to add $20k worth of GPUs and RAM don't bother calling" Pro. The MacBook Air end of the market - where power consumption is far more important than raw performance - was always the low-hanging fruit for Apple Silicon. The Mac Pro isn't really limited by power usage, is very much about adding hulking great GPU cards, and is used by people with very complicated - and expensive-to-change - workflows with large ecosystems of third-party drivers and plug-ins needing to be made compatible. Plus, the current MP only really started shipping in 2020, so Apple are obliged to actively support it for a few years yet - I wouldn't interpret (e.g.) releasing new GPUs as a sign that the Apple Silicon Mac Pros are in trouble. My suspicion is that any AS Mac Pro could - and should - be a bit more radical than a drop-in replacement for the Intel one: e.g. some sort of multiple CPU/GPU/RAM module set-up. Now-ish, while the latest Intel MP is still current and can run in parallel for a while, might be a good time to try that, rather than the past practice of waiting until the former MP is years out of date before trying to foist a shiny new paradigm on users.

Anyhow, we're in - what - month 14 of an "about two years" transition - and what with unavoidable problems like pandemics and global component shortages I think it's a bit early to start calling Apple out for being behind schedule.


Any particular reason to believe that "full" TB4 won't come with the M1X/M2/whatever? Unless you mean "100% glitch and bug free TB4" - which ain't gonna happen on any platform with such a complex and sprawling protocol stack.


Sure, and if the forthcoming higher-end M1X/M2/Whatever MBPs don't support 3 external displays, it would be a bad show. The question is do you need that facility built in to something like a MacBook Air or iPad Pro, or should it be something that warrants buying a higher-end model? It's not just a matter of having the physical ports - you need the CPU and GPU grunt to support it (and that's quite a lot of grunt on a GPU-heavy UI like MacOS that relies on super-high-res screen buffers and scaling to get usable real-estate on 4k displays).

(Or, if you just need lots of text-heavy windows open, and aren't going to be running games, videos or 3D across multi-screens, maybe it's a job for something not bound by physical DisplayPort streams, like Luna Display or DisplayLink...?)

A single external display is going to be perfectly adequate for the traditional MacBook Air (or low-end MBP, which probably only got the "Pro" title because plain "MacBook" was taken) user who got the Air because they value portability over power, and it would be pretty pointless to sacrifice weight and/or battery life to support more (esp. just to earn the TB4 label). Yes, it's a slight downgrade from the Intel versions, but weigh that against all the M1 advantages which partly come from ruthlessly chucking extraneous circuitry off the SoC...

We're in a slightly odd, transitional, place at the moment where the entry-level M1 Macs challenge the higher-end Intel Macs on some aspects of performance, which is extending their appeal beyond the "traditional MacBook Air user" to people who would normally be going for higher-end machines. I think that's a temporary glitch.

It's the same with the whole "who needs an 80Gbps TB5 link" thing - of course some people can make good use of that - the question is, does it need to be integrated into a "mainstream" standard like USB at the expense of 101 exciting new permutations of port capability and cable types, or should it be a new "external PCIe4" port that is a premium feature on high-end PCs? The existence of an 80Gbps interface doesn't magically make CPUs sprout twice the I/O bandwidth - you're going to be looking at Xeon/Epyc-class CPUs (or their future Apple Silicon/ARM/RISC-V equivalents) to do it justice.

You make a lot of fair points.

It is no secret that I have been concerned about Apple's ability to design a full line of processors that can extend into the professional workspace since their transition announcement. Apple's history of transitioning to new processors is littered with a one-step back, two-steps forward mantra, so I am waiting for this upcoming generation of mobile workstations with some anticipation.

I hope you are correct that multiple external display support is coming and I am waiting to see if Apple can indeed implement TB 4 (yes there will likely be glitches) and support multiple external monitors physically or I will need to transition fully back to Windows (something I would rather not do). This is why Apple's lack of support for multiple displays and TB 4 support was disappointing to me.

I agree that this is an odd transitional phase which has been exacerbated by COVID. Perhaps this has given Apple some additional time to develop its future processor lineup. Still I don't have time to wait for Apple to catch up to Intel, much less AMD. At the end of the day, this transition will likely be my last transition as a professional and it is up to Apple to prove their technological prowess or my transition will not be to Apple Silicon but back to Windows.

PS--Thank you for the conversation.

 

[joevt]

I've only come across one, and it's the OWC Thunderbolt Hub. There's no actual benefits to USB 4 afaik, only minor improvements to Thunderbolt 3.

Yes, the OWC Thunderbolt Hub is one I put in the Intel Thunderbolt 4 peripheral category, same as their Thunderbolt 4 dock, and every hub/dock based on the same chip (CalDigit Element Hub, etc.).

No advantage to USB4 except if anyone makes a controller, it won't have the Intel Thunderbolt tax added on. We'll see if that makes a difference when they become available. Maybe someone will make a USB4 device with DisplayPort inputs so people can connect more Thunderbolt displays (this may depend on OS drivers to enable that).

 

[theluggage]

I hope you are correct that multiple external display support is coming and I am waiting to see if Apple can indeed implement TB 4 (yes there will likely be glitches) and support multiple external monitors physically or I will need to transition fully back to Windows (something I would rather not do).

If not, it would be a major disappointment and a failure of Apple Silicon, but it's a no-brainer that the new Pro Macs need to support multiple displays, and I don't think there are any grounds for predicting that when we're only half way through the transition and a massive pandemic-shaped spanner has been thrown into the works. Most of the rumours suggest that the new chips will have upgraded GPUs (which is key to supporting more displays). Also, Apple are now completely in the driving seat - they're not subject to the whims of Intel, Nvidia or AMD (...even in the PPC days they were reliant on Motorola and IBM).

Of course, Apple are also perfectly capable of messing up - the iMac shows that they're still obsessed with form over function - but perhaps its best to wait until we see the glass before we decide if it's half full.

 

[theluggage]

No advantage to USB4 except if anyone makes a controller, it won't have the Intel Thunderbolt tax added on.

Well the ability to use hubs rather than daisy-chaining is new to the USB4/TB4 spec, although it looks like Apple have been able to implement it in software in Big Sur.

As for prices - well, on the host side, AFAIK TB4 is built into 11th gen Intel processors (and AFAIK USB 4 is coming Real Soon Now for Ryzen) so there will be plenty of PCs supporting it. I guess Intel are hoping that USB4's dumpster fire of optional capabilities will become the byword for "plug and pray" and everybody will stump up for TB4 certification.

The question is whether the demand for USB4/TB4 peripherals will take off when the bulk of the mass market is adequately served by USB 3. Economy of scale is probably the most significant factor in the price of electronics - I suspect that the slow uptake of Thunderbolt is mainly because the mass market simply doesn't need it. Even USB 3.1 gen 2 devices are still (a) uncommon and (b) premium-priced and I don't think I've ever heard of a device that uses the USB 3.2 x2 modes.

USB4/TB4 could succeed but still be limited to high(er)-end devices: I somehow doubt that we'll see USB4/TB4 hubs going for $30 any time soon. If nothing else, a TB4 hub needs to be able to supply 15W of power for each port and can only count on one helping of 15W from the host - so ~60W power supplies (if not more, so they can power a large laptop) will be de rigour.

...as someone who doesn't need loadsa TB ports, but could do with a bunch of top-level (i.e. not hub) USB2/3 ports for audio etc. I wonder how things like the OWC and Caldigit Elements hubs perform with USB devices c.f. (a) a regular USB2/3 hub or (b) an TB3-era dock with USB controllers?

 

[joevt]

Well the ability to use hubs rather than daisy-chaining is new to the USB4/TB4 spec, although it looks like Apple have been able to implement it in software in Big Sur.

It even works in EFI. If EFI can enumerate the devices on the Thunderbolt 4 hub, then they'll be usable in Catalina, but hot plug doesn't work for the Thunderbolt 4 devices in Catalina.

As for prices - well, on the host side, AFAIK TB4 is built into 11th gen Intel processors (and AFAIK USB 4 is coming Real Soon Now for Ryzen) so there will be plenty of PCs supporting it. I guess Intel are hoping that USB4's dumpster fire of optional capabilities will become the byword for "plug and pray" and everybody will stump up for TB4 certification.

The question is whether the demand for USB4/TB4 peripherals will take off when the bulk of the mass market is adequately served by USB 3. Economy of scale is probably the most significant factor in the price of electronics - I suspect that the slow uptake of Thunderbolt is mainly because the mass market simply doesn't need it. Even USB 3.1 gen 2 devices are still (a) uncommon and (b) premium-priced and I don't think I've ever heard of a device that uses the USB 3.2 x2 modes.

USB4 is weird. It's basically Thunderbolt 3 but since it's a USB spec they had to add a useless USB tunnelling feature to make it relevant for USB. Useless because every current USB4 host (currently just M1 Macs and Thunderbolt 4 PCs) supports PCIe tunnelling and every current USB4 peripheral (currently just Thunderbolt 4 peripherals) includes a USB controller. It will only be useful when you have cheap hosts that skip out on PCIe tunnelling support (or maybe the user want's to disable PCIe tunnelling for security? but that's what the Thunderbolt 4 DMA security feature is for). The problem with USB tunnelling is that it isn't faster than USB 3.1 gen 2 and it uses the USB controller of the host computer (the M1 Macs have slow USB controllers for 5 Gbps and 10 Gbps connections). USB tunnelling + DisplayPort tunnelling will be an improvement though but only compared to USB-C DisplayPort Alt Mode since the PCIe tunnelling + DisplayPort tunnelling of Thunderbolt 3 has the same benefit. Current USB-C Alt Mode only supports 25.92 Gbps DisplayPort + 480 Mbps USB or 12.96 Gbps DisplayPort + 10 Gbps USB. USB4/TB3 supports 40 Gbps total, up to 35 Gbps for DisplayPort (up to two connections) and whatever remains up to 22 Gbps for USB (using multiple controllers) or any other PCIe data.

USB4/TB4 could succeed but still be limited to high(er)-end devices: I somehow doubt that we'll see USB4/TB4 hubs going for $30 any time soon. If nothing else, a TB4 hub needs to be able to supply 15W of power for each port and can only count on one helping of 15W from the host - so ~60W power supplies (if not more, so they can power a large laptop) will be de rigour.

...as someone who doesn't need loadsa TB ports, but could do with a bunch of top-level (i.e. not hub) USB2/3 ports for audio etc. I wonder how things like the OWC and Caldigit Elements hubs perform with USB devices c.f. (a) a regular USB2/3 hub or (b) an TB3-era dock with USB controllers?

Good questions. I've only seen numbers related to bandwidth. Latency is a different question. There aren't any Intel Macs with Thunderbolt 4 hosts yet so we don't know how much better USB tunnelling will be for them compared to what the M1 Macs have.

Usually devices don't use all the power of a USB port, so those 60W power supplies will only be outputting like 10W or less (if they're not charging a laptop). Just think about how much power a mouse or keyboard uses. A USB audio device will probably have its own power supply. Bus powered storage devices will use the most power but modern SSDs or flash drives don't use much.

 

[theluggage]

Useless because every current USB4 host (currently just M1 Macs and Thunderbolt 4 PCs) supports PCIe tunnelling and every current USB4 peripheral (currently just Thunderbolt 4 peripherals) includes a USB controller.

Useless at the moment when the only USB4 hosts are guaranteed to have PCIe not least because they are still, largely, PCIe bus-based devices. Going forward, though, I can see phones, tablets, embedded and other non-Intel devices - that weren't really a target for Thunderbolt but will be for USB - wanting to implement USB4 without bothering with PCIe. You might even see systems-on-a-chip with just USB4 coming out of the package, rather than the usual mixture of PCIe, DisplayPort and USB 3....

Plus, for peripherals, TB -> USB3 is obviously simpler than TB->PCIe->USB3, especially if you need additional USB 3 ports (the built-in USB controller is a zero-sum game if the peripheral has a daisy-chain port).

These are devices that aren't looking for single, super-fast I/O channels so much as a way to funnel lots of slower peripherals through a single physical port... which, of course, is the old problem with "USB-C" trying to cater for everything from a smart fridge to a supercomputer... and will result in (at least) two distinct and incompatible ways of implementing USB ports on a USB4 peripheral...

The problem with USB tunnelling is that it isn't faster than USB 3.1 gen 2 and it uses the USB controller of the host computer

I expect that USB 3.1 gen 2 will (a) be the end of the road for USB 3.X-based technology and (b) remain the bog-standard "good enough for most things" connectivity method for the next 5-10 years, with USB4/40Gbps peripherals a premium option mostly reserved for high-end displays... with the only really mass-market application being multi-port hubs for multiple slower devices.

Usually devices don't use all the power of a USB port, so those 60W power supplies will only be outputting like 10W or less (if they're not charging a laptop).

True but, if I've got it right, the spec says each USB4 port must be capable of 7.5W (and TB4 requires at least one to be 15W...?) which seems to rule out a purely bus-powered hub actually qualifying as "USB4". Maybe we'll see cheap hubs on fleaBay that play fast and loose with this...

 

[joevt]

Useless at the moment when the only USB4 hosts are guaranteed to have PCIe not least because they are still, largely, PCIe bus-based devices. Going forward, though, I can see phones, tablets, embedded and other non-Intel devices - that weren't really a target for Thunderbolt but will be for USB - wanting to implement USB4 without bothering with PCIe. You might even see systems-on-a-chip with just USB4 coming out of the package, rather than the usual mixture of PCIe, DisplayPort and USB 3....

Plus, for peripherals, TB -> USB3 is obviously simpler than TB->PCIe->USB3, especially if you need additional USB 3 ports (the built-in USB controller is a zero-sum game if the peripheral has a daisy-chain port).

These are devices that aren't looking for single, super-fast I/O channels so much as a way to funnel lots of slower peripherals through a single physical port... which, of course, is the old problem with "USB-C" trying to cater for everything from a smart fridge to a supercomputer... and will result in (at least) two distinct and incompatible ways of implementing USB ports on a USB4 peripheral...


I expect that USB 3.1 gen 2 will (a) be the end of the road for USB 3.X-based technology and (b) remain the bog-standard "good enough for most things" connectivity method for the next 5-10 years, with USB4/40Gbps peripherals a premium option mostly reserved for high-end displays... with the only really mass-market application being multi-port hubs for multiple slower devices.



True but, if I've got it right, the spec says each USB4 port must be capable of 7.5W (and TB4 requires at least one to be 15W...?) which seems to rule out a purely bus-powered hub actually qualifying as "USB4". Maybe we'll see cheap hubs on fleaBay that play fast and loose with this...

Yes, USB tunnelling (with removing PCIe tunnelling) makes doing USB simpler.

USB 3.2 Gen 2x2 is still a possibility for USB tunnelling though none of the current Thunderbolt 4/USB4 controllers (M1, Tiger Lake, Goshen Ridge, Maple Ridge) support it. Gen 2x2 is 20 Gbps (≈ 2100 MB/s - I don't have my own numbers for that yet) which is near the 22 Gbps of Thunderbolt (≈ 2800 MB/s). I don't think macOS supports gen 2x2 yet without a third party driver.

Agreed that unpowered USB4 hubs is probably not a thing.